The effect of feature extraction and data sampling on credit card fraud detection

Abstract Training a machine learning algorithm on class-imbalanced dataset can be difficult task, process that could prove even more challenging under conditions of high dimensionality. Feature extraction and data sampling are among the most popular preprocessing techniques. is used to derive richer set reduced features, while mitigate class imbalance. In this paper, we investigate these two techniques, using credit card fraud four ensemble classifiers (Random Forest, CatBoost, LightGBM, XGBoost). Within context feature extraction, Principal Component Analysis (PCA) Convolutional Autoencoder (CAE) methods evaluated. With regard sampling, Random Undersampling (RUS), Synthetic Minority Oversampling Technique (SMOTE), SMOTE Tomek The F1 score Area Under Receiver Operating Characteristic Curve (AUC) metrics serve as measures classification performance. Our results show implementation RUS method followed by CAE leads best performance for detection.